Physics-data Cooperative Modeling for Ship Motion Prediction

Abstract

Increasing development on autonomous vehicles and concern on ship navigation safety put forward a higher requirement for ship motion forecasting technology. The predictions of ship motion in the near future can give the operator (autonomous ship operating system or human) ample time to respond and avoid dangerous operations. Therefore, modeling and predicting the behavior of ships have been pursued extensively to enhance state estimation and motion control. Vessels operating on the surface of the ocean are exposed to an array of uncertainties, such as the external perturbations produced by wind, waves, and sea currents, etc. Creating an advanced model that comprehensively represents the system and its interaction with its immediate environments has always been challenging. Existing ship motion prediction approaches leverage a wide variety of modeling techniques. The dynamic models can propagate the estimated states into the future. Still, due to the nonlinearity, time-varying dynamics, and coupling with also time-varying environments, it is difficult to derive a good state estimator with high accuracy from observing and understanding the complex system. The fast advancement in instrumentation and data analysis techniques offers an alternative solution by constructing end-to-end models based only on data sampled from ships. For the lack of physical interpretability and inspection of internal structure, these data-based models do not always meet the expectations. The new knowledge and technology to bridge the gap are in demand. The physics-data hybrid concept is an approach of cooperation, differentiation, and maximizing the potential of both models. Either model offers partial solutions for the vessel system. To ensure optimal outcomes, different modeling principles are working in a cooperative way, which requires the capabilities of both segments to operate as efficiently as possible. Leveraging the speed and flexibility advantages of the data-driven technologies and ensuring the robustness and quality of the high-fidelity physics-based model, the cooperative modeling appears to provide fast and accurate predictions for offshore surface vessels. This dissertation exploits the physics-data cooperative modeling methodology and contextualizes the synthesis in maritime motion prediction. In the hybrid framework, modeling principles and formats of fusion aligned differently at various operation scenarios. Three case studies are conducted to validate the developed physics-data cooperative models for optimization and prediction. The first one relates to physics-based disciplines and enabling applications. The rest deploy two forms of cooperation. Experiments are carried out in both simulator and the vessel R/V Gunnerus operating in the real world. The results confirm the enhanced mode quality and prediction performance observed for the physics-data cooperative models.